Increasing macroscalar instruction level parallelism

Abstract

A processor may include a vector functional unit that supports concurrent operations on multiple data elements of a maximum element size. The functional unit may also support concurrent execution of multiple distinct vector program instructions, where the multiple vector instructions each operate on multiple data elements of less than the maximum element size.

Description

BACKGROUND[0001]1. Technical Field[0002]This disclosure relates to processors, and more particularly to increasing instruction level parallelism for vector instructions.[0003]2. Description of the Related Art[0004]Processor performance can be increased by increasing the degree to which computational work can be performed concurrently. Performing multiple distinct computational tasks in parallel may not result in either task finishing more quickly—in fact, under some circumstances, the design considerations necessary to support concurrency may decrease the performance of individual tasks relative to their best-case performance. But even despite a possible performance penalty for individual tasks, concurrent processing of multiple tasks may increase the aggregate amount of processing work done per cycle, thus improving overall computational performance.[0005]Vector processing represents one approach to increasing concurrency. Generally speaking, vector instructions may operate on mult...

AI Technical Summary

Benefits of technology

[0006]Various embodiments of a method for increasing the instruction level parallelism of vector program instructions such as Macroscalar instructions are disclosed. Broadly speaking, a processor is contemplated in which a functional unit provides vector execution resources that are capable of operating on vectors of data elements having a maximum element size. In instances where multiple vector program instructions do not require the maximum element size, the multiple instructions may execute concurrently within the vector execution resources, increasing utilization of the resources that would otherwise have been underutilized if the multiple instructions had been executed serially.

Problems solved by technology

Performing multiple distinct computational tasks in parallel may not result in either task finishing more quickly—in fact, under some circumstances, the design considerations necessary to support concurrency may decrease the performance of individual tasks relative to their best-case performance.

However, if the data elements to be processed do not require the full complement of execution resources implemented by a processor, the excess resources may be underutilized, and thus effectively wasted.

Images